Integrate persistent SQLite indexes with native server reads

[DandyLyons]

Goal

Replace repeated collection discovery with a shared native indexing service used directly by the CLI and server. Part of Phase 2.5 in #93; depends on #138, #139, and #140.

Requirements

• Integrate indexed filesystem reads behind existing repository/RecordStore boundaries; keep files authoritative.
• Publish consistent refreshed data snapshots without unexpectedly changing the immutable EndpointPlan.
• Define explicit behavior for resource configuration changes and failed refreshes.
• Add pagination, filtering, and limits with deterministic behavior; expose full-text search only when enabled under Add optional body indexing with single-copy FTS5 storage #146. Declare search capability in configuration and the generated OpenAPI contract, with explicit behavior when unavailable.
• Generate new read parameters/responses from the same plan and test OpenAPI parity.
• Preserve identity collisions, overlapping membership, and visibility of invalid rule-selected records.
• Indexing a rule/type never automatically exposes an HTTP resource.
• Use direct library calls rather than spawning the CLI.
• Consume Bound collection refresh memory and reuse extraction across scopes #147's bounded refresh/publication service and Use runtime-selected JSONB with a JSON-text compatibility mode #148's encoding policy. Do not introduce ripgrep subprocesses into server reads.
• Preserve Build the immutable server read snapshot and generic record representation #89's body-bearing record representation in metadata-only mode through bounded authoritative-file reads, checked against indexed revisions. Specify refresh or explicit failure when files change; never silently combine old indexed metadata with a newly read different body version.
• Bound page materialization and response buffering. Do not rebuild a corpus-sized in-memory body cache in immutable read snapshots. Preserve the existing RecordStore and EndpointPlan boundaries.

Acceptance criteria

• CLI and server reuse indexing semantics and persistence.
• Concurrent reads see consistent generations as updates occur.
• Existing server contract fixtures retain their semantics.
• Pagination/filter/search and generated OpenAPI agree.
• Refresh failures and configuration changes are explicit and tested.
• Metadata-only and FTS-enabled resources expose their actual search capabilities consistently with OpenAPI.
• Metadata-only body reads preserve record/revision consistency under concurrent file changes.
• Large collections and body-bearing pages use bounded materialization rather than caching the entire corpus in memory.

Boundaries

This is an indexed filesystem adapter, not a canonical SQLiteRecordStore. Portable Core and the later Workers runtime do not require native SQLite extensions. UI editing is outside #93.